Natural Language Processing

A complete natural language processing system should receive incoming speech through microphones which convert the sounds into electrical signals. These are analysed and processed by the system. The system forms some sort of software model which represents an understanding of the input. It generates an appropriate response, which is converted into speech-like sound output by a loudspeaker.

Learning Intentions and Success Criteria

Learning Intentions

Success Criteria

In this tutorial I am going to learn :
  • how to ...
By the end of the tutorial I will be able to :
  • ...

Speech Recognition

The purpose of speech recognition is to convert incoming sound waves into a list of words. Note that this stage may not be needed in some NLP systems, where the input is in typed or even written form.

Speech recognition is not an easy process. Speech is not usually in the form of discrete words.

Firstly, the output from the microphone must be converted into digital data. This is done by analysing the signal using a frequency spectrograph. This provides a digital analysis of the sound.

Next, the continuous data stream must be broken up into individually identifiable sounds.

The frequency spectrogram produced by the incoming sound is split up into sound fragments which can be matched to a library of phonemes. This is not an exact process, and ambiguities can arise at this stage, just as it can with human hearing: " Did she say "witch" or "which"? " If the sound quality is not very good, a th may sound more like a f. Any background noises may also be picked up and interpreted as phonemes.

Natural Language Understanding

This is the purpose of natural language understanding, which of course, applies to both spoken and written / typed input.
NLU is commonly split into 3 stages. These are:
  • Syntactic analysis
  • Semantic analysis
  • Pragmatic analysis

Syntactic analysis involves checking that the words fit together into allowed structures, like sentences. At its simplest, it may use pattern matching against standard sentence patterns. This method was used by Eliza, one of the earliest natural language research programs. However, simple pattern matching is not very useful, except in very limited circumstances. A more effective method is to use grammars and parsing.

Semantic analysis is about extracting the meaning from a sentence. For example, "the boy ate the chocolate" is a valid sentence in English in terms of syntactic analysis. It also makes sense. However, "the flower drove the house" is also a valid sentence in terms of syntactic analysis, but it makes no sense at all. Semantic analysis takes a syntactically correct sentence and uses various methods to represent the meaning (if any) of the sentence.

Pragmatic analysis involves considering a sentence in its context. This may include surrounding sentences, or some background knowledge about the person speaking or the situation. Also, a sentence may not mean what it appears to mean. For example, the question "Can you shut the door?" isn't really a question - it is really an instruction! These type of issues are dealt with during the pragmatic analysis stage.

Speech generator

You need to know this is the 3rd stage but nothing about it

Speech Synthesis

You need to know this is the 4th stage but nothing about it


Here are the building blocks of Grammer :


Exam Questions

Embed Document here


  • Please make a list of essential key-points here


  • upload files here